[Feature] Add the ability to work with images

[nxexox]

Problem

I can only create applications that take already arrays of primitives and return primitive structures. But if I want to send a picture, I need to preprocess it into an array of numbers and send it. It would be cool to add data transformers for images and vice versa. This will allow the use of models that accept images as input and models that produce images as output.

Examples

Config file:

...
ml:
    data_transformer_for_predict: "mlup.ml.data_transformers.image_data_transformer.ImageJPGDataTransformer"
    data_transformer_for_predicted: "mlup.ml.data_transformers.image_data_transformer.ImageJPGDataTransformer"
web:
    inner_data_type: "file". # Default is primitive

request.post("http://0.0.0.0:8009/predict", files={'upload_file': open('image.jpg', 'rb')})